Related Experiment Video
Updated: Mar 29, 2026
13:01
Industrialized, Artificial Intelligence-guided Laser Microdissection for Microscaled Proteomic Analysis of the Tumor Microenvironment
Published on: June 3, 2022
4.7K
Direct Segmentation of Mammography and Tomosynthesis Sinograms for Lesion Localization
Estefanía Ruíz Muñoz1, Leopoldo Altamirano Robles1, Raquel Díaz Hernández1
1Instituto Nacional de Astrofísica, Óptica y Electrónica, Sta. María Tonantzintla, Puebla 72840, Mexico.
Tomography (Ann Arbor, Mich.)
|March 27, 2026
Summary
Directly segmenting sinograms improves breast lesion detection and localization in mammography and digital breast tomosynthesis (DBT). This novel approach enhances accuracy, especially for challenging cases, offering a clinically viable solution.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Breast lesion detection and localization are challenging in mammography and digital breast tomosynthesis (DBT) due to tissue overlap and reconstruction artifacts.
- Sinograms offer a potential solution by preserving original projection data, avoiding reconstruction-induced information loss.
Purpose of the Study:
- To develop and evaluate a direct segmentation approach for breast lesions using mammography and DBT sinograms.
- To assess the efficacy of a U-Net architecture for sinogram-based lesion detection and localization.
Main Methods:
- A U-Net architecture was employed for direct segmentation of mammography and DBT sinograms.
- Experiments utilized the CBIS-DDSM and Breast Cancer Screening DBT datasets, with patient-level data splitting to ensure robust evaluation.
- Three input configurations were tested: mammography sinograms, DBT sinograms, and a combined multimodal model.
Main Results:
- The mammography sinogram model achieved high performance (Dice: 0.90 on test set), significantly outperforming the DBT sinogram model (Dice: 0.70).
- Multimodal fusion enhanced DBT results (Dice: 0.84), and centroid analysis demonstrated high correspondence with annotations (99.11%).
- The proposed method showed superior lesion localization compared to YOLOv5x, particularly for small or multiple lesions.
Conclusions:
- Direct sinogram segmentation is an effective and clinically applicable strategy for breast lesion detection and localization.
- This approach overcomes limitations of traditional reconstruction-based methods, offering improved accuracy and efficiency.
